AI Engineering Breakthrough Slashes Project Failure Rates

Engineered Intelligence: A New Approach to AI Implementation: The concept of “engineered intelligence” is emerging as a potential solution to the high failure rate of AI projects and the looming threat of another AI winter.

• The current approach to AI implementation often involves data scientists attempting to engineer real-world solutions, resulting in an 87% failure rate for AI projects.
• Engineered intelligence aims to create a distinct discipline for applied artificial intelligence, similar to how other scientific breakthroughs are handed off to specialized engineers for practical application.

The Problem with Current AI Implementation: The lack of a dedicated field for applying AI research to real-world solutions has led to a mismatch between expertise and application.

• Data scientists, trained primarily in theoretical AI research, are often tasked with developing practical AI solutions, leading to inefficiencies and failures.
• This approach contrasts with other scientific disciplines where laboratory breakthroughs are typically handed off to specialized engineers for real-world application.

Defining Engineered Intelligence: This emerging discipline focuses on the practical application of AI research, rooted in engineering principles and methodologies.

• Engineered intelligence aims to leverage AI breakthroughs alongside raw materials to design and build safe, practical value.
• The approach allows domain experts, scientists, and engineers to create intelligence solutions without needing to become data scientists themselves.

Industry Adoption and Potential Benefits: Leading industrial organizations are beginning to incorporate engineered intelligence into their AI strategies.

• Companies are reestablishing research-to-engineering pipelines and forming new partnerships with academia and technology vendors.
• This shift is creating ecosystemic conditions for AI research to be effectively handed off to intelligence engineers, similar to how chemical research is shared with chemical engineers.
• The result is breakthrough applications in tangible use cases that create value, make it into production, and might not have been discovered by data scientists or technology vendors based on data alone.

Implementing Engineered Intelligence: The article outlines a five-step approach for introducing intelligence engineering to organizations, contrasting it with the traditional AI implementation method.

• Traditional approach (87% failure rate):

  1. Create a list of problems or examine data
  2. Pick potential use cases
  3. Analyze use cases for ROI, feasibility, cost, and timeline
  4. Choose and invest in a subset of use cases

• Intelligence engineering approach:

  1. Create a heatmap of expertise across existing processes
  2. Assess and score the value and scarcity of expertise
  3. Choose the top five most valuable and scarce expertise areas
  4. Analyze for ROI, feasibility, cost, and timeline to engineer intelligent solutions
  5. Select and invest in a subset of value cases

The Role of Expertise in Intelligence Engineering: The article emphasizes the importance of practical experience and domain knowledge in successful AI implementation.

• Expertise is described as the heart of intelligence engineering, expressed through skills acquired through practical application.
• While theory and training can accelerate skill acquisition, practical experience is deemed essential for developing true expertise.

Future Implications and Potential Impact: The adoption of engineered intelligence could have far-reaching effects on organizations, industries, and society as a whole.

• As organizations and educational institutions build programs for engineered intelligence, it could create a new class of jobs and usher in a wave of value creation.
• The approach has the potential to unlock the unrealized economic and societal benefits of AI by focusing on practical, safe, and valuable applications.

Critical Analysis: Balancing Optimism with Realism: While engineered intelligence presents a promising approach to AI implementation, it’s important to consider potential challenges and limitations.

• The success of this approach will depend on effective collaboration between AI researchers and engineers, which may require significant cultural and organizational changes.
• While the concept addresses important issues in AI implementation, it’s crucial to maintain realistic expectations about the pace and scope of improvements in AI project success rates.
• Further research and real-world case studies will be necessary to fully validate the effectiveness of the engineered intelligence approach across different industries and use cases.